Although there have been notable advances in child survival, the rate of infection-related hospital fatalities remains high among pediatric patients in resource-limited settings. Hence, the critical need for enhancing child survival demands immediate attention. Present risk prediction tools for in-hospital complications have been developed through statistical modelling methods and exhibit specific constraints. This study sought to create and compare an array of machine learning-based models using admission-recorded information to predict all-cause in-patient death in two cohorts of children in Uganda: those under 6 months and those aged between 6 and 60 months. 190 (7.0%) out of 2,698 children under six months and 164 (4.3%) out of 3,835 children over six months of age died following admission. For each cohort, five supervised machine learning algorithms were trained and internally validated on 67% and 33% of the data, respectively. The models incorporated demographics, clinical variables, and photoplethysmography-extracted features as inputs. The balanced random forest classifier demonstrated superior performance on the test set in both cohorts. In the first cohort, the model achieved a sensitivity of 0.95, specificity of 0.71, positive predictive value (PPV) of 0.15, and negative predictive value (NPV) of 0.99. Comparing this to the model that exclusively utilized demographic and clinical parameters, an increase of 14.5% was observed in sensitivity, while specificity and PPV experienced reductions of 4% and 11.8%, respectively. NPV, however, remained unchanged. In the second cohort, the model's sensitivity was 0.75, with a specificity of 0.73, a PPV of 0.08, and an NPV of 0.99. Upon training the model with only demographic and clinical variables in this cohort, the corresponding metrics were 0.67, 0.76, 0.09, and 0.98, respectively. By integrating photoplethysmography data with machine learning algorithms, it may be possible to develop predictive models that can identify high-risk patients who are more likely to experience adverse outcomes like in-hospital death. Such models could empower healthcare providers in ill-equipped settings to allocate limited resources effectively and deliver targeted interventions to those most in need.

We propose an efficient method to label the frames of a video consistently and quickly. Our method is motivated by welding videos, where molten metal moves continuously, and some frames may be noisy due to the high intensity of arc and spatters that can obscure the desired point to be annotated. The traditional annotation methods which pause the videos and annotate each frame individually are time-consuming and may result in inconsistent labeling between different annotators for noisy frames of a video. Our new proposed method benefits from the fact that video sequences are contiguous. We track the location of the cursor while the video is being played, with the user controlling the playback speed, including the playback direction. We process the recorded cursor locations to convert them to the final annotation. Compared to frame-to-frame annotation methods, we show that our proposed interface speeds up the annotating process by 21 times while maintaining consistency of the labels.

Stethoscope-recorded chest sounds provide the opportunity for remote cardio-respiratory health monitoring of neonates. However, reliable monitoring requires high-quality heart and lung sounds. This paper presents novel artificial intelligence-based Non-negative Matrix Factorisation (NMF) and Non-negative Matrix Co-Factorisation (NMCF) methods for neonatal chest sound separation. To assess these methods and compare them with existing single-channel separation methods, an artificial mixture dataset was generated comprising heart, lung, and noise sounds. Signal-to-noise ratios were then calculated for these artificial mixtures. These methods were also tested on real-world noisy neonatal chest sounds and assessed based on vital sign estimation error, and a signal quality score of 1-5, developed in our previous works. Overall, both the proposed NMF and NMCF methods outperform the next best existing method by 2.7 dB to 11.6 dB for the artificial dataset, and 0.40 to 1.12 signal quality improvement for the real-world dataset. The median processing time for the sound separation of a 10 s recording was found to be 28.3 s for NMCF and 342 ms for NMF. With the stable and robust performance of our proposed methods, we believe these methods are useful to denoise neonatal heart and lung sounds in the real-world environment.

With advances in digital stethoscopes, internet of things, signal processing and machine learning, chest sounds can be easily collected and transmitted to the cloud for remote monitoring and diagnosis. However, low quality of recordings complicates remote monitoring and diagnosis, particularly for neonatal care. This paper proposes a new method to objectively and automatically assess the signal quality to improve the accuracy and reliability of heart rate (HR) and breathing rate (BR) estimation from noisy neonatal chest sounds. A total of 88 10-second long chest sounds were taken from 76 preterm and full-term babies. Six annotators independently assessed the signal quality, number of detectable beats, and breathing periods from these recordings. For quality classification, 187 and 182 features were extracted from heart and lung sounds, respectively. After feature selection, class balancing, and hyperparameter optimization, a dynamic binary classification model was trained. Then HR and BR were automatically estimated from the chest sound and several approaches were compared.The results of subject-wise leave-one-out cross-validation, showed that the model distinguished high and low quality recordings in the test set with 96% specificity, 81% sensitivity and 93% accuracy for heart sounds, and 86% specificity, 69% sensitivity and 82% accuracy for lung sounds. The HR and BR estimated from high quality sounds resulted in significantly less median absolute error (4 bpm and 12 bpm difference, respectively) compared to those from low quality sounds. The methods presented in this work, facilitates automated neonatal chest sound auscultation for future telehealth applications.

Verifying safety of closed-loop drug delivery systems is a crucial step to obtain regulatory approval for such devices. Recently, we proposed a safety-preserving platform to formally verify the safety of closed-loop propofol anesthesia in the presence of uncertainty in patient models. This platform verifies that closed-loop anesthesia maintains the propofol concentration within the therapeutic window. To improve safety for at-risk patients, additional safety constraints on physiological variables, such as blood pressure, can be taken into account. To do so, an accurate description of the propofol effect on blood pressure as well as a comprehensive characterization of inter-patient variability are required. In this paper, we identify and validate a set of models characterizing the blood pressure response to propofol for 10 individual patients from clinical data. This model set shows uncertainty in the effect of propofol on blood pressure among at-risk patients. We use this model set to verify the safety of closed-loop propofol anesthesia subject to constraints on blood pressure using our safety verification platform. We show that there exists an admissible dose of propofol that maintains the propofol plasma concentration as well as blood pressure within safe bounds, without the need for additional treatment for hypotension. However, the dose of propofol which maintains safety, may not provide sufficient anesthesia for all patients and in critical cases treatment for hypotension would be required.

Situation awareness (SA) is essential for the safe operation of systems involving human-automation interaction. In this paper, using the theory of functional observers, we model SA for the user interacting with a continuous-time linear time-invariant dynamical system. For systems under human control or shared control, we use the proposed model to determine the required information to be displayed in the user interface for achieving SA. The user interface provides the user with the ability to observe the continuous-time outputs of the system, as well as the ability to enter continuous-time control inputs. In some systems, due to inadequacy of the displayed information, the user may not be able to accomplish the desired task. To determine the required information to be displayed and the necessary states to be tracked, we propose a model of attaining SA for the users by modeling the user as a specific type of estimator (i.e., the extended delayed functional observer/predictor). We then evaluate what information is needed for such an estimator and how the desired functional of the states have to be expanded so that the user can precisely reconstruct and accurately predict the desired task. As an application example, we investigate the problem of controlling the depth of anesthesia during surgery and determine whether there exists a feasible combination of the expanded task and the displayed information that allows the anesthetist to precisely predict the depth of anesthesia of the patient.

We present a study evaluating two respiratory rate estimation algorithms using videos obtained from placing a finger on the camera lens of a mobile phone. The two algorithms, based on Smart Fusion and empirical mode decomposition (EMD), consist of previously developed signal processing methods to detect features and extract respiratory induced variations in photoplethysmographic signals to estimate respiratory rate. With custom-built software on an Android phone, photoplethysmographic imaging videos were recorded from 19 healthy adults while breathing spontaneously at respiratory rates between 6 to 32 breaths/min. Signals from two pulse oximeters were simultaneously recorded to compare the algorithms' performance using mobile phone data and clinical data. Capnometry was recorded to obtain reference respiratory rates. Two hundred seventy-two recordings were analyzed. The Smart Fusion algorithm reported 39 recordings with insufficient respiratory information from the photoplethysmographic imaging data. Of the 232 remaining recordings, a root mean square error (RMSE) of 6 breaths/min was obtained. The RMSE for the pulse oximeter data was lower at 2.3 breaths/min. RMSE for the EMD method was higher throughout all data sources as, unlike the Smart Fusion, the EMD method did not screen for inconsistent results. The study showed that it is feasible to estimate respiratory rates by placing a finger on a mobile phone camera, but that it becomes increasingly challenging at respiratory rates greater than 20 breaths/min, independent of data source or algorithm tested.

The development of mobile applications for the diagnosis and management of pregnant women with pre-eclampsia is described. These applications are designed for use by community-based health care providers (c-HCPs) in health facilities and during home visits to collect symptoms and perform clinical measurements (including pulse oximeter readings). The clinical data collected in women with pre-eclampsia are used as the inputs to a predictive model providing a risk score for the development of adverse outcomes. Based on this risk, the applications provide recommendations on treatment, referral, and reassessment. c-HCPs can access patient records across multiple visits, using multiple devices that are synchronized using a secure Research Electronic Data Capture server. A unique feature of these applications is the ability to measure oxygen saturation with a pulse oximeter connected to a smartphone (Phone Oximeter). The mobile health application development process, including challenges encountered and solutions are described.

A modular framework for the development of medical applications that promotes deterministic, robust and correct code is presented. The system is based on the portable Gambit Scheme programming language and provides a flexible cross-platform environment for developing graphical applications on mobile devices as well as medical instrumentation interfaces running on embedded platforms. Real world applications of this framework for mobile diagnostics, telemonitoring and automated drug infusions are reported. The source code for the core framework is open source and available at: https://github.com/part-cw/lambdanative.

For Linear Time-Invariant (LTI) systems under human control or under shared control, we develop a technique to determine the required information content of the display - that is, the device which provides the user with information about the system. Having the right information on the display can help the user to attain better understanding about the system and apply required inputs to control it. To determine the required information to be displayed, we first model the user as a specific type of observer with certain understanding regarding the derivatives of the input and output signals. We then evaluate what information has to be provided for such an observer to make the desired states of the system observable. We provide an example on modeling the pilot of a Boeing B-747 during the phase of approach to landing. In this example we also determine the required information in the display to perform a desired task.

We present a connection between the viability kernel and maximal reachable sets. Current numerical schemes that compute the viability kernel suffer from a complexity that is exponential in the dimension of the state space. In contrast, extremely efficient and scalable techniques are available that compute maximal reachable sets. We show that under certain conditions these techniques can be used to conservatively approximate the viability kernel for possibly high-dimensional systems. We demonstrate the results on two practical examples, one of which is a seven-dimensional problem of safety in anesthesia.

This paper presents a systematic approach to system identification and closed-loop control of end-tidal carbon dioxide partial pressure (PETCO2) in mechanically ventilated patients. An empirical model consisting of a linear dynamic system followed by an affine transform is proposed to derive a low-order and high-fidelity representation that can reproduce the positive and inversely proportional dynamic input-output relationship between PETCO2 and minute ventilation (MV) in mechanically ventilated patients. The predictive capability of the empirical model was evaluated using experimental respiratory data collected from eighteen mechanically ventilated human subjects. The model predicted PETCO2 response accurately with a root-mean-squared error (RMSE) of 0.22+/-0.16 mmHg and a coefficient of determination (r2) of 0.81+/-0.18 (mean+/-SD) when a second-order rational transfer function was used as its linear dynamic component. Using the proposed model, a closedloop control method for PETCO2 based on a proportionalintegral (PI) compensator was proposed by systematic analysis of the system root locus. For the eighteen mechanically ventilated patient models identified, the PI compensator exhibited acceptable closed-loop response with a settling time of 1.27+/- 0.20 min and a negligible overshoot (0.51+/-1.17%), in addition to zero steady-state PETCO2 set point tracking. The physiologic implication of the proposed empirical model was analyzed by comparing it with the traditional multi-compartmental model widely used in pharmacological modeling.

This paper presents an analytical approach to design a continuous time-invariant two-level control scheme for asymptotic stabilization of a desired period-one trajectory for a hybrid model describing walking by a planar biped robot with noninstantaneous double-support phase and point feet. It is assumed that the hybrid model consists of both single- and double-support phases. The design method is based on the concept of hybrid zero dynamics. At the first level, parameterized continuous within-stride controllers, including single- and double-support-phase controllers, are employed. These controllers create a family of 2-D finite-time attractive and invariant submanifolds on which the dynamics of the mechanical system is restricted. Moreover, since the mechanical system during the double-support phase is overactuated, the feedback law during this phase is designed to be minimum norm on the desired periodic orbit. At the second level, parameters of the within-stride controllers are updated by an event-based update law to achieve hybrid invariance, which results in a reduced-order hybrid model for walking. By these means, stability properties of the periodic orbit can be analyzed and modified by a restricted Poincaré return map. Finally, a numerical example for the proposed control scheme is presented.

Alerting systems used by current physiological monitors are designed to detect changes in the levels of vital signs, but they tend to be very sensitive to artifacts. This paper proposes a method to detect changes in the direction of trend and generate multilevel alerts according to the statistical significance of the detection. One-point-ahead signal predictions are calculated by averaging the historical data with the weights decreasing in the past. The two-sided cumulative sums (Cusum) of the prediction errors are tested against multiple thresholds to detect change points with two levels of certainty. The temporal shapes of the detected changes are analyzed using heuristics to determine whether to trigger an alert. The method was tested offline using 20 cases collected during surgery at a local hospital. The detection results were evaluated by two experienced anesthesiologists. The direction of trend was correctly detected in 90.2% of the annotated changes for end-tidal carbon dioxide, 89.4% for expiratory minute volume, 91.8% for peak airway pressure, and 95.4% for noninvasive blood pressure. The certainty levels of the true-positive alerts estimated by the algorithm have a high ratio of agreement with the anesthesiologists' evaluations.

In this paper, we propose two encoding parameters to facilitate the design of a rhythm-based tactile scheme for humans. The sense of touch has been used for many years to aid communication for people with sensory impairments. Now, vibrations are used in mobile phones and handheld computers but are generally very basic and do not fully exploit the potential of vibration as a means of communication. Several studies explored the use of tactile icons combining different parameters such as amplitude modulation, location, and duration. However, the parameter ldquorhythmrdquo has not attracted a lot of attention. Using two parameters (muE, sigmaE) to control the design of tactile-pattern sets, we created four stimulus schemes of 20 patterns each. Using a tactile belt located around the waist, 64 subjects tested them. The amount of static information transferred and response times (RTs) varied among the different designs. The scheme with the best set of encoding parameters (muE=3.sigmaE>0) conveyed 4 b of information or 16 tokens with an RT of 4.3 s.

We developed a novel wearable tactile display system as an alternative to the visual and audio displays routinely used by anesthesiologists to monitor patients in the operating room (OR). Visual displays and auditory alarms can be distracting or insufficient in their alarm transmission whereas a tactile display, which utilizes the sense of touch, can act as an effective conduit for alert delivery. A sophisticated alarm scheme is essential to convey the complex array of physiological information available in current monitoring systems; therefore, to report all relevant alerts to the attending anesthesiologist, it is essential that an augmenting or replacement display system be at least as effective and efficacious as conventional systems. Using multidimensional Tactons, we designed a tactile alert scheme consisting of 36 unique stimuli and evaluated the accuracy and response time in stimuli recognition using a tactile prototype worn as a belt. We observed an overall accuracy of 81% and a response time of 4.8 seconds. 4.18 bits (18.07 tokens) of messages were successfully communicated without loss of information. These results demonstrate that the novel tactile display represents an effective and potentially work-load-reducing method to convey vital information non-visually and non-aurally.

This study investigates a low-pass circuit featuring saturation. The describing function method is used to linearize the model. One commonly encountered low-pass device is the RLC filter, which is frequently used in power electronic applications to smooth the wave forms of the circuit variables. However, in case the values of the variables pertaining to the inductance exceed those of the rated, the corner frequency of the filter may change drastically. This is due to the nonlinear phenomenon ferroresonance. A systematic procedure for the analysis is presented that with reasonable accuracy accommodates most inductors represented by nonlinear magnetization curves. The results of the study show the importance of considering saturation in the design process of the aforementioned filter and how the bandwidth of the same is influenced by the circuit parameters

This paper presents a new blind equalization algorithm called multimodulus algorithm (MMA). This algorithm combines the benefits of the well-known reduced constellation algorithm (RCA) and constant modulus algorithm (CMA). In addition, MMA provides more flexibility than RCA and CMA, and is better suited to take advantage of the symbol statistics of certain types of signal constellations, such as nonsquare constellations, very dense constellations, and some wrong solutions.

Different methods of searching for dynamic neural models with minimum complexity have been proposed. The performance as well as the optimality of such methods highly depend on the way "model complexity" is defined. On the other hand, the learning theory creates a framework to assess the learning properties of models. These properties include the required size of the training samples as well as the statistical confidence over the model. In this paper, we first apply the learning properties of the reciprocal multi-quadratic radial basis function networks to introduce a new measure of complexity, which provides a balance between the training and testing performances of the model. Then, we present a systematic evolutionary programming technique that searches for a neural model of an unknown system with the optimal structure as well as parameters. The performance of the novel evolutionary method is illustrated by a numerical modeling simulation that testifies to the success of the proposed method.

It is shown how genetic algorithms can be applied for system identification of both continuous and discrete time systems. It is shown that they are effective in both domains and are able to directly identify physical parameters or poles and zeros. This can be useful because changing one physical parameter might affect every parameter of a system transfer function. The estimates of poles and zeros are then used to design a discrete time pole placement adaptive controller. Simulations for minimum and nonminimum phase systems and a system with unmodeled dynamics are presented.>